Research map / Detection and segmentation / CNN features
D2Det: Towards High Quality Object Detection and Instance Segmentation
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use dense local regression instead of traditional box regression (ablation) | Faster R-CNN baseline · MS COCO minival | AP: 38.0 -> 41.5 | Helped |
| Use dense local regression instead of traditional box regression (ablation) | Faster R-CNN baseline · MS COCO minival | AP@0.75: 41.5 -> 44.8 | Helped |
| Add binary overlap prediction to dense local regression (ablation) | Dense local regression · MS COCO minival | AP: +0.7 AP | Helped |
| Use discriminative RoI pooling instead of baseline pooling (ablation) | Faster R-CNN baseline · MS COCO minival | AP: 38.0 -> 39.3 | Helped |
| Add adaptive weighting to discriminative RoI pooling (ablation) | Discriminative RoI pooling · MS COCO minival | AP: +0.4 AP | Helped |
| Combine dense local regression and discriminative RoI pooling (ablation) | Faster R-CNN baseline · MS COCO minival | AP: 38.0 -> 42.7 | Helped |
| Use dense local regression rather than grid-guided keypoint localization (ablation) | Grid R-CNN Plus · MS COCO minival | AP: 40.2 -> 41.5 | Helped |
| Use D2Det instead of Cascade R-CNN | Two-stage detection on MS COCO · MS COCO test-dev | AP: 42.8 -> 45.4 | Helped |
| Use D2Det instead of DCN v2 with the same backbone | ResNet101-deform v2 detector · MS COCO test-dev | AP: 44.0 -> 47.4 | Helped |
| Use multi-scale training and inference | D2Det · MS COCO test-dev | AP: 47.4 -> 50.1 | Helped |
| Use D2Det instead of HTC | Instance segmentation · MS COCO test-dev | Mask AP: 39.7 -> 40.2 | Helped |
| Use D2Det instead of HTC | Instance segmentation · MS COCO test-dev | inference time (ms): 339 -> 168 | Helped |
Closest papers by meaning
- DALDet: Depth-Aware Learning Based Object Detection for Autonomous Driving (AAAI 2024 · similarity 0.85)
- RemDet: Rethinking Efficient Model Design for UAV Object Detection (AAAI 2025 · similarity 0.85)
- Union-over-Intersections: Object Detection beyond Winner-Takes-All (ICLR 2025 · similarity 0.84)
- D-FINE: Redefine Regression Task of DETRs as Fine-grained Distribution Refinement (ICLR 2025 · similarity 0.83)
- Solving Instance Detection from an Open-World Perspective (CVPR 2025 · similarity 0.83)
- DI-MaskDINO: A Joint Object Detection and Instance Segmentation Model (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.